Explaining Artificial Intelligence to my Son | Reinforcement Learning

Explaining Artificial Intelligence to my Son | Reinforcement Learning

January 24, 2020 3 By Stanley Isaacs


[Music] Ooookay baby! Time for Baba to sleep which means time for baby to sleep. Maybee a bedtime story? How about reinforcement learning? Say Baba wants to teach his Roomba how to cross the room. The robot needs to get from one end of the room to the other really fast. It can just meander around and finally get to the other side like it normally does but if we start to score how well it did then using *Reinforcement Learning* it can learn to go faster. Think of the Roomba instead as a mouse we’ll call him Mousy. Mousy is in a maze that looks just like the room and on the other end there’s some nice stinky cheese, Mousy’s favorite! Mousy wants to get to the cheese fast so he can eat it but he doesn’t know the best way to get to the cheese. It can start just like the Roomba, just meandering around until it sees the cheese when he’s done we can rate his attempt and give him a score. If he wastes too much time or goes back to where he came from we give him a small score and Mousy learns to avoid doing the things that he did that time. If he gets straight to the point and gets to the cheese fast we give a large score and Mousy gets more encouraged about the things that he did Now mousey has a terrible memory so it takes him a long time to learn sometimes he just meanders around to to explore but eventually he learns!!! And that is exactly what reinforcement learning is The robot is just like mousey we tell it to do something and then rate what it did when it finishes maybe give it a low score if it takes too long to get to the end and it slowly will learn to avoid those paths. Kind of like how babies learn too… but only if they listen to their Baba!!! 🙂 Alright come on baby let’s put you to bed. Do you want another story? How about Bayesian Statistics? [Music]